xref: /freebsd/sys/kern/kern_thr.c (revision 3c6e15bceeab4470243c60c9a4b5b9cafca9abaa) 
1 /*-
2  * Copyright (c) 2003, Jeffrey Roberson <jeff@freebsd.org>
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice unmodified, this list of conditions, and the following
10  *    disclaimer.
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in the
13  *    documentation and/or other materials provided with the distribution.
14  *
15  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
16  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
17  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
18  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
19  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
20  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
21  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
22  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
23  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
24  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
25  */
26 
27 #include <sys/cdefs.h>
28 __FBSDID("$FreeBSD$");
29 
30 #include "opt_compat.h"
31 #include "opt_posix.h"
32 #include <sys/param.h>
33 #include <sys/kernel.h>
34 #include <sys/lock.h>
35 #include <sys/mutex.h>
36 #include <sys/priv.h>
37 #include <sys/proc.h>
38 #include <sys/posix4.h>
39 #include <sys/resourcevar.h>
40 #include <sys/sched.h>
41 #include <sys/sysctl.h>
42 #include <sys/smp.h>
43 #include <sys/syscallsubr.h>
44 #include <sys/sysent.h>
45 #include <sys/systm.h>
46 #include <sys/sysproto.h>
47 #include <sys/signalvar.h>
48 #include <sys/ucontext.h>
49 #include <sys/thr.h>
50 #include <sys/rtprio.h>
51 #include <sys/umtx.h>
52 #include <sys/limits.h>
53 
54 #include <machine/frame.h>
55 
56 #include <security/audit/audit.h>
57 
58 #ifdef COMPAT_IA32
59 
60 extern struct sysentvec ia32_freebsd_sysvec;
61 
62 static inline int
63 suword_lwpid(void *addr, lwpid_t lwpid)
64 {
65 	int error;
66 
67 	if (curproc->p_sysent != &ia32_freebsd_sysvec)
68 		error = suword(addr, lwpid);
69 	else
70 		error = suword32(addr, lwpid);
71 	return (error);
72 }
73 
74 #else
75 #define suword_lwpid	suword
76 #endif
77 
78 extern int max_threads_per_proc;
79 extern int max_threads_hits;
80 
81 static int create_thread(struct thread *td, mcontext_t *ctx,
82 			 void (*start_func)(void *), void *arg,
83 			 char *stack_base, size_t stack_size,
84 			 char *tls_base,
85 			 long *child_tid, long *parent_tid,
86 			 int flags, struct rtprio *rtp);
87 
88 /*
89  * System call interface.
90  */
91 int
92 thr_create(struct thread *td, struct thr_create_args *uap)
93     /* ucontext_t *ctx, long *id, int flags */
94 {
95 	ucontext_t ctx;
96 	int error;
97 
98 	if ((error = copyin(uap->ctx, &ctx, sizeof(ctx))))
99 		return (error);
100 
101 	error = create_thread(td, &ctx.uc_mcontext, NULL, NULL,
102 		NULL, 0, NULL, uap->id, NULL, uap->flags, NULL);
103 	return (error);
104 }
105 
106 int
107 thr_new(struct thread *td, struct thr_new_args *uap)
108     /* struct thr_param * */
109 {
110 	struct thr_param param;
111 	int error;
112 
113 	if (uap->param_size < 0 || uap->param_size > sizeof(param))
114 		return (EINVAL);
115 	bzero(&param, sizeof(param));
116 	if ((error = copyin(uap->param, &param, uap->param_size)))
117 		return (error);
118 	return (kern_thr_new(td, &param));
119 }
120 
121 int
122 kern_thr_new(struct thread *td, struct thr_param *param)
123 {
124 	struct rtprio rtp, *rtpp;
125 	int error;
126 
127 	rtpp = NULL;
128 	if (param->rtp != 0) {
129 		error = copyin(param->rtp, &rtp, sizeof(struct rtprio));
130 		if (error)
131 			return (error);
132 		rtpp = &rtp;
133 	}
134 	error = create_thread(td, NULL, param->start_func, param->arg,
135 		param->stack_base, param->stack_size, param->tls_base,
136 		param->child_tid, param->parent_tid, param->flags,
137 		rtpp);
138 	return (error);
139 }
140 
141 static int
142 create_thread(struct thread *td, mcontext_t *ctx,
143 	    void (*start_func)(void *), void *arg,
144 	    char *stack_base, size_t stack_size,
145 	    char *tls_base,
146 	    long *child_tid, long *parent_tid,
147 	    int flags, struct rtprio *rtp)
148 {
149 	stack_t stack;
150 	struct thread *newtd;
151 	struct proc *p;
152 	int error;
153 
154 	error = 0;
155 	p = td->td_proc;
156 
157 	/* Have race condition but it is cheap. */
158 	if (p->p_numthreads >= max_threads_per_proc) {
159 		++max_threads_hits;
160 		return (EPROCLIM);
161 	}
162 
163 	if (rtp != NULL) {
164 		switch(rtp->type) {
165 		case RTP_PRIO_REALTIME:
166 		case RTP_PRIO_FIFO:
167 			/* Only root can set scheduler policy */
168 			if (priv_check(td, PRIV_SCHED_SETPOLICY) != 0)
169 				return (EPERM);
170 			if (rtp->prio > RTP_PRIO_MAX)
171 				return (EINVAL);
172 			break;
173 		case RTP_PRIO_NORMAL:
174 			rtp->prio = 0;
175 			break;
176 		default:
177 			return (EINVAL);
178 		}
179 	}
180 
181 	/* Initialize our td */
182 	newtd = thread_alloc();
183 	if (newtd == NULL)
184 		return (ENOMEM);
185 
186 	/*
187 	 * Try the copyout as soon as we allocate the td so we don't
188 	 * have to tear things down in a failure case below.
189 	 * Here we copy out tid to two places, one for child and one
190 	 * for parent, because pthread can create a detached thread,
191 	 * if parent wants to safely access child tid, it has to provide
192 	 * its storage, because child thread may exit quickly and
193 	 * memory is freed before parent thread can access it.
194 	 */
195 	if ((child_tid != NULL &&
196 	    suword_lwpid(child_tid, newtd->td_tid)) ||
197 	    (parent_tid != NULL &&
198 	    suword_lwpid(parent_tid, newtd->td_tid))) {
199 		thread_free(newtd);
200 		return (EFAULT);
201 	}
202 
203 	bzero(&newtd->td_startzero,
204 	    __rangeof(struct thread, td_startzero, td_endzero));
205 	bcopy(&td->td_startcopy, &newtd->td_startcopy,
206 	    __rangeof(struct thread, td_startcopy, td_endcopy));
207 	newtd->td_proc = td->td_proc;
208 	newtd->td_ucred = crhold(td->td_ucred);
209 
210 	cpu_set_upcall(newtd, td);
211 
212 	if (ctx != NULL) { /* old way to set user context */
213 		error = set_mcontext(newtd, ctx);
214 		if (error != 0) {
215 			thread_free(newtd);
216 			crfree(td->td_ucred);
217 			return (error);
218 		}
219 	} else {
220 		/* Set up our machine context. */
221 		stack.ss_sp = stack_base;
222 		stack.ss_size = stack_size;
223 		/* Set upcall address to user thread entry function. */
224 		cpu_set_upcall_kse(newtd, start_func, arg, &stack);
225 		/* Setup user TLS address and TLS pointer register. */
226 		error = cpu_set_user_tls(newtd, tls_base);
227 		if (error != 0) {
228 			thread_free(newtd);
229 			crfree(td->td_ucred);
230 			return (error);
231 		}
232 	}
233 
234 	PROC_LOCK(td->td_proc);
235 	td->td_proc->p_flag |= P_HADTHREADS;
236 	newtd->td_sigmask = td->td_sigmask;
237 	thread_link(newtd, p);
238 	bcopy(p->p_comm, newtd->td_name, sizeof(newtd->td_name));
239 	thread_lock(td);
240 	/* let the scheduler know about these things. */
241 	sched_fork_thread(td, newtd);
242 	thread_unlock(td);
243 	if (P_SHOULDSTOP(p))
244 		newtd->td_flags |= TDF_ASTPENDING | TDF_NEEDSUSPCHK;
245 	PROC_UNLOCK(p);
246 	thread_lock(newtd);
247 	if (rtp != NULL) {
248 		if (!(td->td_pri_class == PRI_TIMESHARE &&
249 		      rtp->type == RTP_PRIO_NORMAL)) {
250 			rtp_to_pri(rtp, newtd);
251 			sched_prio(newtd, newtd->td_user_pri);
252 		} /* ignore timesharing class */
253 	}
254 	TD_SET_CAN_RUN(newtd);
255 	sched_add(newtd, SRQ_BORING);
256 	thread_unlock(newtd);
257 
258 	return (error);
259 }
260 
261 int
262 thr_self(struct thread *td, struct thr_self_args *uap)
263     /* long *id */
264 {
265 	int error;
266 
267 	error = suword_lwpid(uap->id, (unsigned)td->td_tid);
268 	if (error == -1)
269 		return (EFAULT);
270 	return (0);
271 }
272 
273 int
274 thr_exit(struct thread *td, struct thr_exit_args *uap)
275     /* long *state */
276 {
277 	struct proc *p;
278 
279 	p = td->td_proc;
280 
281 	/* Signal userland that it can free the stack. */
282 	if ((void *)uap->state != NULL) {
283 		suword_lwpid(uap->state, 1);
284 		kern_umtx_wake(td, uap->state, INT_MAX, 0);
285 	}
286 
287 	PROC_LOCK(p);
288 	sigqueue_flush(&td->td_sigqueue);
289 	PROC_SLOCK(p);
290 
291 	/*
292 	 * Shutting down last thread in the proc.  This will actually
293 	 * call exit() in the trampoline when it returns.
294 	 */
295 	if (p->p_numthreads != 1) {
296 		thread_stopped(p);
297 		thread_exit();
298 		/* NOTREACHED */
299 	}
300 	PROC_SUNLOCK(p);
301 	PROC_UNLOCK(p);
302 	return (0);
303 }
304 
305 int
306 thr_kill(struct thread *td, struct thr_kill_args *uap)
307     /* long id, int sig */
308 {
309 	struct thread *ttd;
310 	struct proc *p;
311 	int error;
312 
313 	p = td->td_proc;
314 	error = 0;
315 	PROC_LOCK(p);
316 	if (uap->id == -1) {
317 		if (uap->sig != 0 && !_SIG_VALID(uap->sig)) {
318 			error = EINVAL;
319 		} else {
320 			error = ESRCH;
321 			FOREACH_THREAD_IN_PROC(p, ttd) {
322 				if (ttd != td) {
323 					error = 0;
324 					if (uap->sig == 0)
325 						break;
326 					tdsignal(p, ttd, uap->sig, NULL);
327 				}
328 			}
329 		}
330 	} else {
331 		if (uap->id != td->td_tid)
332 			ttd = thread_find(p, uap->id);
333 		else
334 			ttd = td;
335 		if (ttd == NULL)
336 			error = ESRCH;
337 		else if (uap->sig == 0)
338 			;
339 		else if (!_SIG_VALID(uap->sig))
340 			error = EINVAL;
341 		else
342 			tdsignal(p, ttd, uap->sig, NULL);
343 	}
344 	PROC_UNLOCK(p);
345 	return (error);
346 }
347 
348 int
349 thr_kill2(struct thread *td, struct thr_kill2_args *uap)
350     /* pid_t pid, long id, int sig */
351 {
352 	struct thread *ttd;
353 	struct proc *p;
354 	int error;
355 
356 	AUDIT_ARG(signum, uap->sig);
357 
358 	if (uap->pid == td->td_proc->p_pid) {
359 		p = td->td_proc;
360 		PROC_LOCK(p);
361 	} else if ((p = pfind(uap->pid)) == NULL) {
362 		return (ESRCH);
363 	}
364 	AUDIT_ARG(process, p);
365 
366 	error = p_cansignal(td, p, uap->sig);
367 	if (error == 0) {
368 		if (uap->id == -1) {
369 			if (uap->sig != 0 && !_SIG_VALID(uap->sig)) {
370 				error = EINVAL;
371 			} else {
372 				error = ESRCH;
373 				FOREACH_THREAD_IN_PROC(p, ttd) {
374 					if (ttd != td) {
375 						error = 0;
376 						if (uap->sig == 0)
377 							break;
378 						tdsignal(p, ttd, uap->sig, NULL);
379 					}
380 				}
381 			}
382 		} else {
383 			if (uap->id != td->td_tid)
384 				ttd = thread_find(p, uap->id);
385 			else
386 				ttd = td;
387 			if (ttd == NULL)
388 				error = ESRCH;
389 			else if (uap->sig == 0)
390 				;
391 			else if (!_SIG_VALID(uap->sig))
392 				error = EINVAL;
393 			else
394 				tdsignal(p, ttd, uap->sig, NULL);
395 		}
396 	}
397 	PROC_UNLOCK(p);
398 	return (error);
399 }
400 
401 int
402 thr_suspend(struct thread *td, struct thr_suspend_args *uap)
403 	/* const struct timespec *timeout */
404 {
405 	struct timespec ts, *tsp;
406 	int error;
407 
408 	error = 0;
409 	tsp = NULL;
410 	if (uap->timeout != NULL) {
411 		error = copyin((const void *)uap->timeout, (void *)&ts,
412 		    sizeof(struct timespec));
413 		if (error != 0)
414 			return (error);
415 		tsp = &ts;
416 	}
417 
418 	return (kern_thr_suspend(td, tsp));
419 }
420 
421 int
422 kern_thr_suspend(struct thread *td, struct timespec *tsp)
423 {
424 	struct timeval tv;
425 	int error = 0, hz = 0;
426 
427 	if (tsp != NULL) {
428 		if (tsp->tv_nsec < 0 || tsp->tv_nsec > 1000000000)
429 			return (EINVAL);
430 		if (tsp->tv_sec == 0 && tsp->tv_nsec == 0)
431 			return (ETIMEDOUT);
432 		TIMESPEC_TO_TIMEVAL(&tv, tsp);
433 		hz = tvtohz(&tv);
434 	}
435 
436 	if (td->td_pflags & TDP_WAKEUP) {
437 		td->td_pflags &= ~TDP_WAKEUP;
438 		return (0);
439 	}
440 
441 	PROC_LOCK(td->td_proc);
442 	if ((td->td_flags & TDF_THRWAKEUP) == 0)
443 		error = msleep((void *)td, &td->td_proc->p_mtx, PCATCH, "lthr",
444 		    hz);
445 	if (td->td_flags & TDF_THRWAKEUP) {
446 		thread_lock(td);
447 		td->td_flags &= ~TDF_THRWAKEUP;
448 		thread_unlock(td);
449 		PROC_UNLOCK(td->td_proc);
450 		return (0);
451 	}
452 	PROC_UNLOCK(td->td_proc);
453 	if (error == EWOULDBLOCK)
454 		error = ETIMEDOUT;
455 	else if (error == ERESTART) {
456 		if (hz != 0)
457 			error = EINTR;
458 	}
459 	return (error);
460 }
461 
462 int
463 thr_wake(struct thread *td, struct thr_wake_args *uap)
464 	/* long id */
465 {
466 	struct proc *p;
467 	struct thread *ttd;
468 
469 	if (uap->id == td->td_tid) {
470 		td->td_pflags |= TDP_WAKEUP;
471 		return (0);
472 	}
473 
474 	p = td->td_proc;
475 	PROC_LOCK(p);
476 	ttd = thread_find(p, uap->id);
477 	if (ttd == NULL) {
478 		PROC_UNLOCK(p);
479 		return (ESRCH);
480 	}
481 	thread_lock(ttd);
482 	ttd->td_flags |= TDF_THRWAKEUP;
483 	thread_unlock(ttd);
484 	wakeup((void *)ttd);
485 	PROC_UNLOCK(p);
486 	return (0);
487 }
488 
489 int
490 thr_set_name(struct thread *td, struct thr_set_name_args *uap)
491 {
492 	struct proc *p = td->td_proc;
493 	char name[MAXCOMLEN + 1];
494 	struct thread *ttd;
495 	int error;
496 
497 	error = 0;
498 	name[0] = '\0';
499 	if (uap->name != NULL) {
500 		error = copyinstr(uap->name, name, sizeof(name),
501 			NULL);
502 		if (error)
503 			return (error);
504 	}
505 	PROC_LOCK(p);
506 	if (uap->id == td->td_tid)
507 		ttd = td;
508 	else
509 		ttd = thread_find(p, uap->id);
510 	if (ttd != NULL)
511 		strcpy(ttd->td_name, name);
512 	else
513 		error = ESRCH;
514 	PROC_UNLOCK(p);
515 	return (error);
516 }
517